The Hermite-Hadamard inequality in higher dimensions

From MaRDI portal
Publication:2303799



Abstract: The Hermite-Hadamard inequality states that the average value of a convex function on an interval is bounded from above by the average value of the function at the endpoints of the interval. We provide a generalization to higher dimensions: let OmegasubsetmathbbRn be a convex domain and let f:OmegaightarrowmathbbR be a convex function satisfying , then frac{1}{|Omega|} int_{Omega}{f ~d mathcal{H}^n} leq frac{2 pi^{-1/2} n^{n+1}}{|partial Omega|} int_{partial Omega}{f~d mathcal{H}^{n-1}}. The constant 2pi−1/2nn+1 is presumably far from optimal, however, it cannot be replaced by 1 in general. We prove slightly stronger estimates for the constant in two dimensions where we show that 9/8leqc2leq8. We also show, for some universal constant c>0, if OmegasubsetmathbbR2 is simply connected with smooth boundary, f:OmegaightarrowmathbbR is subharmonic, i.e. Deltafgeq0, and , then int_{Omega}{f~ d mathcal{H}^2} leq c cdot mbox{inradius}(Omega) int_{partial Omega}{ f ~dmathcal{H}^{1}}. We also prove that every domain OmegasubsetmathbbRn whose boundary is 'flat' at a certain scale delta admits a Hermite-Hadamard inequality for all subharmonic functions with a constant depending only on the dimension, the measure |Omega| and the scale delta.


The classical Hermite-Hadamard inequality states that if \(f:[a,b]\to\mathbb{R}\) is convex, then \[\frac{1}{b-a} \int_a^b f(x)\, \mathrm{d} x \leq \frac{f(a)+f(b)}{2}.\] In the present paper, the author establishes a generalization of this result to higher dimensions. Therefore, let \(\Omega\subset \mathbb{R}^n\) be a convex domain and let \(f:\Omega\to\mathbb{R}\) be a convex function. If \(f\big\vert_{\partial \Omega} \geq 0\), then \[\frac{1}{|\Omega|} \int_{\Omega} f \, \mathrm{d} \mathcal{H}^n \leq \frac{c_n}{|\partial \Omega|} \int_{\partial \Omega} f \, \mathrm{d} \mathcal{H}^{n-1}\] where \(c_n=2\pi^{-1/2} n^{n+1}\). In order to prove this result, the Lebesgue measure on \(\Omega\) is in a certain way transported to a measure on \(\partial \Omega\). Note, that the constant \(c_n\) is not optimal. However, the author points out that due to an example by \textit{P. Pasteczka} [Ann. Univ. Paedagog. Crac., Stud. Math. 233(17), 37--43 (2018; Zbl 1428.26065)], the inequality fails in general for \(c_n=1\). Furthermore, the optimal constant \(\bar{c}_n\) needs to satisfy \[\bar{c}_n \leq \left(1+\frac{1}{4n}+o(n^{-1}) \right) n^{n+1}\] as well as \(9/8 \leq \bar{c}_2 \leq 8\). Similar inequalities are established for subharmonic functions on simply connected domains in dimension 2 and convex domains in dimension \(n\geq 3\). Moreover, a Hermite-Hadamard-type inequality is established for subharmonic functions on smooth domains which are ``sufficiently flat.



Cites work









This page was built for publication: The Hermite-Hadamard inequality in higher dimensions

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q2303799)